Virtual agent
4.3.5 Openness and flexibility
Openness to further exploration and interpretation (i.e. sense-making), and flexible structure are two interrelated characteristics of an EM model (cf. §4.1). In practical EM, the openness and flexibility of EM model are realised with the use of an interpreter and definitive notations in the tkeden modelling environment. In contrast to the use of compilers, where the program (or model) have to be compiled and/or encrypted into some form of low level human unreadable text, the use of an interpreter means that the script has to be open source74 in principle. This makes the script always open to change.
Figure 4.13 – The difference between definitive notation and procedural programming
The virtue of a definitive notation is that the order of the definitions has no significance to the state of the model. As a simple example, figure 4.13 illustrates this character of definitive notation with two fragments of definitive scripts and two code fragments in procedural programming. The use of definitive notations in practical EM gives the modeller a flexible structure to construe a situation, in which, new definitions can be added to the definitive scripts and redefinitions can be made on the fly, or at any time during the modelling process. This can be done through the interactive prompt in tkeden modelling environment (cf. §4.3.1 and §4.3.4). Furthermore, the modeller can interact with the definitive script at the definition
74
The term “open source” is here used in its literal sense to refer to the situation where the source code of a program (or model) is unencrypted and available to examination and modification by anybody who is interested.
level – a lower level of abstraction in contrast to, e.g. an object-oriented programming paradigm75, where the data structure (or data model) and the methods for manipulating the data are often encapsulated in the same object. In such a case, it is not easy to make changes to individual data structure without affecting their common parents (i.e. classes). Indeed, the flexible structure in definitive notation eases the exploratory and experimental model-building activities in EM.
4.3.6 Experimentation
As mentioned at the beginning of this chapter, experimentation is one of the key motives for EM. Beynon and Russ (2008) argue that EM provides a holistic conceptual framework for exploratory experiment. In the journal paper “Experimenting with Computing”, Beynon and Russ (2008) distinguish two different kinds of experimental activity: post-theory and exploratory. In post-theory experiment, the context for observation is usually stable, and the outcomes can readily be observed when the key observables are changed. There is often a reliable basis for predicting criterion for successful outcome, whether drawn from theory or previous experience (Beynon and Russ, 2008). Despite the fact that there is some degree of human intelligence involved in setting up an experiment of this such kind, human interpretation is often a “marginal element” (Beynon and Russ, 2008). In contrast, the context for observation in exploratory experiment is preliminary and provisional, and human engagement in this context is crucial.
Beynon and Russ (2008) argue that it is in the context of exploratory experiment that the term ‘experiment’ has its authentic meaning, namely, “taking an action whose effect is unknown”. They distinguish such a context sharply from the context for post-theory experiment. In the context of exploratory experiment:
75
The only exception is when the statement is enclosed in a procedure in the EDEN notation, in which case the modeller may have to redefine the entire procedure. This is due to the fact that EDEN is a hybrid notation that consists of both definitive and procedural constructs. However, there is no such restriction in other definitive notations such as DoNaLD or SCOUT.
“… the effect of an action cannot possibly be exactly predicted – indeed, it cannot even be uncontroversially identified or comprehensively registered.” (Beynon and Russ, 2008, p.477)
Consequently, the support for exploratory experimentation is not straight forward in contrast to the support for post-theory experimentation, where the experiment can be set up with reference to suitable formalised procedures when certain key observables have been identified (Beynon and Russ, 2008).
Taken together, the various characteristics of tkeden discussed in the previous sub-sections (cf. §4.3.2 ~ 4.3.5) can be seen as supporting the use of EM for experimentation. Figure 4.14 depicts how these techniques and approaches support different characteristics of practical EM and how they give support for exploratory experimentation using EM.
Arguably, programming paradigms such as procedural, functional, and object-oriented fail to offer the qualities that definitive notations potentially afford for exploratory experimentation. For instance, procedural programming does not allow flexible structure (cf. §4.3.4); object- oriented programming maintains a high level of abstraction which may restrict effective interaction (cf. §4.3.4) and flexible structure (cf. §4.3.5); functional programming offers a poor correspondence between everyday observables and variables in the program which makes interaction and comprehension difficult (cf. Beynon and Russ, 2008).
Figure 4.14 – Potential influence between instant feedback, dialogical interaction, experimentation, openness and flexibility
Experimentation not only needs support from programming paradigms, but also the support from the tool (or environment) as well. For instance, Javascript is an interpretable language, but it is very hard, if not impossible, to experiment with the script in the web browsers because they do not normally provide support for such interactions76. By contrast, spreadsheet environments provide much better support. A spreadsheet potentially offers instant feedback, openness and flexibility for creating numerical models. However, it does not provide an interaction history so that the modeller has no idea how he has reached the current situation.
In summary, I argue that current EM practice with the tkeden modelling environment potentially offers better support to exploratory experimentation. In the next chapter, I will discuss how EM can be practiced in collaborative modelling, an activity which I claim is central to groupware development (also cf. chapter 7).
76
An interactive prompt is not a common feature and is not usually available in web browsers. Furthermore, the javascript console, if provided, is merely for displaying error messages, and not for taking input from the developer.